Suicide Attempts Among Formerly Hospitalized Adolescents: A Prospective Naturalistic Study of Risk During the First 5 Years After Discharge

Suicide Attempts Among Formerly Hospitalized Adolescents: A Prospective Naturalistic Study of Risk During the First 5 Years After Discharge DAVID B. GOLDSTON , PH.D., STEPHANIE SERGENT DANIEL, PH.D.. DAVID M. REBOUSSIN, PH.D., BETH A. REBOUSSIN, PH.D., PATRICIA H. FRAZIER. M.A., AND ARTHUR E. KELLEY, M.D.

ABSTRACT Objective: To examine risk for suicide attempts among 180 consecutively referred adolescents during the first 5 years after discharge from an inpatient psychiatry unit. Method: In a prospective naturalist ic study, adolescents were assessed at psychiatric hospitalizat ion and semiannually thereafter for up to 5 years with semistructured psychiatric diagnostic interviews and self-report quest ionnaires . Results: Approximately 25% of the adolescents attempted suicide and no adolescents completed suicide within the first 5 years after discharge. The first 6 months to 1 year after discharge represented the period of highest risk . The number of prior attempts was the strongest predictor of posthospitalization attempts. Affect ive disorders by themselves did not predict later suicide attempts but were related to posthosp italization attempts when accompanied by a history of past suicide attempts. Independent of psychiatric diagnoses, severity of depressive symptoms and trait anxiety also predicted suicide attempts . Similar to the effect with affective disorders, depressive symptoms were most strongly related to posthospitalization suicidality among adolescents with a prior history of suicide attempts . Conclusions: Particularly among youths with prior suicidal behavior , clinicians should be alert to the above constellation of psychiatric predictors of posthospitalization suicidal behavior. J. Am. Acad. Child Adolesc. Psychiatry. 1999. 38(6):66CHl71. Key Words: longitudinal study, hosp italization , adolescence. suicidal behavior.

Among adolescents, it has been suggested that future research and preventive effort s in suicidal behavio r focus on "high-risk" groups (Rorheram-Borus, 1989). There are 2 reasons for thi s suggestion. First, certain high-ri sk groups account for a disproportionate number of adolescent suicide attempts. Second, primary prevention effort s targeted widely at the gener al adolescent population have not proved efficacious in reducing the occurrence of suicidal behavior (Shaffer er al., 1990) , in part

()8')O-8S(, 7/ ')') /J8()(,-O (,(,OC l) I')')') by the Am erican Academ y of C hi ld and Ad olescent Psych iatry.

because of the low base rate of the targeted beh avior, suicide attempts. More focused prevention and intervention efforts might prove comparatively more efficacious , more time- and more cost-efficient. In addition , studies focusing on high -risk groups may yield more informative results about purported predictors of suicide attempts because of the higher base rates of suicide attempts in these sam ples. One group of adolescents thought to be at especially high risk for later suicidal behavior is youths who have been psychiatrically hospitalized (e.g., Barter et al., 1968; Kerfoot and McHugh, 1992; King er al., 1995; Mclntire et al., 1977). However, the degree to which these youths arc at risk for attempting suicide at any given point after hospitalization is not clear. Few studies have been designed to precisely estimate or delineate periods of risk by examining suicide att empts at repeated po ints in time (e.g., Spirito et al., 1992) or over an extended period of time after hospitalization (e.g., Angle et al., 1983; Pfeffer et al., 1993). Moreover, the origins of the risk associated with hospit alization are not well understood. It remains an empirical question as to whether it is the pr ior history of

660

J . AM. ACAD. CHILD AD O LE SC. PSYCHIATRY. .18:6 , JU N E 1999

Accrplfd Nurembrr 4. /998. Drs. Goldstan, Dani rl, and A'rl/l')' 'l11d M« . Frazier are with thr Department of P~ydJi,ury and BehavioralMrdicille. alld Dr>. D. Rrbaussin and H. Reboussi» orr with the Department of Public Hraltb Scimcrs, m ,ke Form University Schuol of Mrdirinr, Wimto,,-S~z/rm. N C Thisprojrct uiasfimdrd I~y NIMH [milt MH48762 and bya Faculty Scholar Award from tb« W'il/iam 'I.' Gmllt Foundationto Dr. Goldston. Theauthorsth'l11k Richard W. Brunstettrr; M'D.. Cindy Danner. M.S. "'t,'ry,ot Homan Holloman, Ph.D.. AllryRl'f(istrr. M.A.. Katbenn« Kirkhart. M.A.. IlIId S. LYII Tre,ui,l'llY. M.A..

w..

f or the ir assistance wid} this stlu'.y.

or

Correspondenre 10 Dr. Goldston. Departmrnt P~ychio lry and H.ehol'ioml Medicine, Iffikr Form University SdlOol ofMedicinr. Medica! Centrr BOIII""lrd, Wimto 'I-.~~"rm . NC 271.57-1087.

ADOLESCENT SUICIDE ATTEMPTS

suicide attempts, the presence of psychiatric disorder, or some other factor intrinsic to hospitalized youths that portends this heightened risk. The recency of past suicidal behavior and the presence of suicidal ideation among adolescents without recent attempts also have not been closelyexamined as predictors of posthospitalization risk. In clinical practice, hospitalized youths who voice suicidal ideation or have made recent suicide attempts are often considered to be at highest risk for posthospiralization suicide attempts. Nonetheless, the predictive utility of suicidal ideation among youths with histories of suicide attempts has often not been examined. In addition, many studies of inpatient suicidal adolescents have focused only on youths with recent suicide attempts or suicidal behavior which precipitated hospitalizations, precluding examination of whether the temporal proximity of the suicidal behavior is related to later risk. Among psychiatric risk factors, the relationship between psychiatric disorder and suicidal behavior has received the greatest attention. For example, cross-sectional studies have suggested that suicidal adolescents, including suicide cornplerers, have high rates of affective disorders (Andrews and Lewinsohn, 1992; Brent et aI., 1993b). Indeed, Shaffer et al. (1996) found that affective disorder was associated with risk for youth suicide even after controlling for prior history of suicide attempts. Follow-up studies of preadolescent youths have also indicated that most suicide attempts occur during or proximal to episodes of mood disorders (Kovacset al., 1993; Pfeffer et al., 1993) and suggest that early-onset depression is predictive of suicidal behavior during adolescence and adulthood (Harrington et al., 1994; Kovacs et al., 1993). Nonetheless, given the complexity and severity of problems among all youths in inpatient settings and the fact that psychiatric hospitalization in and of itself is a risk factor for suicidal behavior (Hawton et aI., 1993), it is not clear whether affective disorders among inpatient youths have the same prognostic utility as might be the case in other samples. It also is not clear whether affective disorders among hospitalized youths are predictive of later suicide attempts after controlling for history of past suicidal behavior and sociodemographic variables. Nondepressive disorders such as conduct disorders and substance use disorders also have been implicated in cross-sectional case-control and community studies as psychiatric risk factors for suicidal ideation and attempted and completed suicide (Andrews and Lewinsohn, 1992;

Beautrais et aI., 1998; Brent et aI., 1993b; Gould et aI., 1998; Shaffer et aI., 1996). Longitudinal studies, however, have not consistently found conduct disorder to be an independent risk factor for later suicide attempts (Brent et al., 1993a; Kovacs et aI., 1993; Pfeffer et al., 1993). Some data, in fact, have indicated that it may be the co morbidity of affective disorders and conduct or substance abuse disorders, rather than conduct or substance abuse disorder in isolation, that is associated with increased risk for suicidal behavior (e.g., Brent et aI., 1993a; Kovacs et aI., 1993). Severity of psychiatric symptomatology also has been thought to be associated with risk for suicide attempts. For example, cross-sectional studies have described adolescent suicide attempters in hospital settings (Ohring et al., 1996), and repeat attempters in particular (Goldston et aI., 1996), as evidencing significantly more trait anxiety than nonsuicidal peers. Repeat suicide attempters also report more severe depressive symptoms on selfreport questionnaires than other hospitalized youths (Goldston et al., 1996). Myers et al. (1991b) found severity of anger from the K-SADS interview to be 1 of 3 factors associated with later suicidality among adolescents. However, we know of no studies of hospitalized adolescents that have investigated whether severity of symptomatology is predictive of suicidal behavior after controlling for the presence of psychiatric disorder and past suicide attempts. Finally, studies have not examined the differential utility of psychiatric variables in predicting suicide attempts among youths with and without a history of suicidal behavior. Psychiatric variables may be more predictive of later suicidal behavior among individuals who already have demonstrated the capability of attempting suicide than among individuals without such histories. However, longitudinal studies have not examined whether psychiatric variables are differentially predictive of future suicidal behavior depending on the history of suicide attempts. The present study is a prospective, naturalistic, repeated assessment investigation of 180 consecutively referred adolescents after hospitalization. The study was specifically designed to examine the risk for suicidal behavior over time in this high-risk sample. Several questions are addressed in this article. First, are there high-risk periods of time after hospitalization for suicide attempts? Second, does the risk over time for posthospitalization suicide attempts differ for adolescents with and without histories of past attempts? Third, does recency of past sui-

J. AM. ACAD. CHILD ADOLESC. PSYCHIATRY, .J8:6, JUNE I')')')

661

GOLDSTON ET AI..

cidal behavior or the presence of suicidal ideation predict which youths are at higher risk for attempting suicide after hospitalization? Fourth, to what degree are affective disorders, conduct or oppositional disorders , substance use disorders, and/or psychiatric comorbidity associated with later suicide attempts? Fifth, to what degree are severity of depressive symptomatology, anxiety, and anger associated with later suicid aliry, after controlling for the presence of psychiatric disord er and history of suicidal behavior? And sixth, are psychiatric predictors of suicid e attempts differentially predictive among previously suicidal and previously nonsuicidal adolescents? Each of these questions is addressed while the potentially moderating effects of age, gender, and race are considered. METHOD Subjects and Overview of Procedure The 180 adolescent s parti cipating in this study were hospitalized between September 4, 1991. and April 10 . 1995. The youths were recruited from among co nsecutive d ischarges from th e Adol escent Inpatient Psych iatry Unit of Wake Forest University Baptist Medical C ente r. Sam pling was not based o n prior histo ry of suicidal beh avior. To be eligible for the study. ado lescents needed to meet the following inclusionary criteria: (I) ages 12 to 19 years, (2) no evidence of mental retard ation , (3) ad m ission to th e unit for at least 10 days, (4) no evidence of serio us systemic physical d isease such as insulin-dependent diabetes mellitus or seizure d isord er, (5) still residing in North C arolina or Virgin ia at the time of the first follow-up assessment. (6) not a sibling of a subject already parti cipatin g in the stu dy. and (7) able to coope rate with and com plete the inpatie nt assessment. Youths eligible for the study becau se they were hospitalized for at least 10 days did not differ from youth s with sho rter stays in gender and race CO Illposition or age at hospitalization. However. youths with longe r Stays did report more severe depressive sympto ms than youths with shorter Stays (mean Beck Depression Invent or y [BDI] score = 13.9 for Stays < 10 days; mean BDI score = 19.6 for stays ~ 1O days; t = 1.98.p = .049). T he subjects in this longitudinal study were recruited from a larger sam ple of 26 9 adolescents described at hospit alization in anothe r publ ication (Go ldsto n er al., 1998). To recru it the planned sample of 180 . we attempted to locate 22 5 yout hs one-half year after the ir hospitalization. One subject died (of cardiac problems ) before he could be asked to pa rt icipate. We were able to find 96 .0% of the remaining pool of eligible subjects; of th ese. 83 .7% agreed to part icipate in the lon gitud inal study. Adolescents wh o agr eed to particip at e in th e follow-up stu dy and adolescents who decli ned d id not differ in age. gend er, or racial distribution ; in whether they had attempted suicide with in 2 weeks preceding th eir index hosp italization or had made suicide att empts previously; o r in their mean BDI, State-Trait Anxiety Inventory (STAll Trait Anxiety. or State-Trait Anger/Anger Expression Inventory (STAXI) Trait Anger scores. The sam ple for the present study co nsisted of9 1 girls (50.6%) and 89 boys (49.4%) . who se ages ran ged from 12 to 18 years (mea n = 14 .8 years. med ian = 14.8 years) at the tim e of hospitalizatio n. The majority of the su bjects (80 %) wer e wh ite. 16.7% were African Am erican. and the remainder were Hi span ic. Nati ve-Ameri can, or of Asian-American heritage. At stu dy ent ry, 54% lived with at least one

662

biological parent; 18% lived with both biological parents. Sixteen pe rcent of yo ut hs were in custody of the Department of Social Services. As classified by th e H ollingshead (1957) Inde x. th e socioeconomi c status (SES) of youths not in custody of social services was d istributed as follows: I (highest), 3.3% ; II, 12.6 % ; III , 21. 9% ; IV. 29 .8% ; V (lowest), 32 .4% . For stat istical analyses, SES was dichotomized as high (I, II, III) versus low (IV. V). All yout hs admitted to the inpatient psychiat ry unit were assessed during their hospitalizati on with a sta ndard ized batter y including semistruc tu red interview instruments. T hereafter. yout hs were asked to parti cipate in semian nual follow-up assessments for up to 5 years. Approximately hal f of the se inte rviews took place in the med ical cent er; the remainder were scheduled off-site in subjects' hom es. residential treatment faciliti es. hosp itals, group homes, prisons, and training schools. The interviews were typically scheduled every 6 to 8 months but varied within and among subjects because of scheduling conflicts, staff shortages , a nd subject requests. To maximize th e resources of the study, youth s who ent ered the study at the beginning of subject recruitment were followed for a longer period of tim e th an youths who ente red the study at the end of the recruitment period . At the cuto ff date for these analyses (April 22 . 1997), subjects in the study had been followed from 0.5 year to 5.5 year (mean = 3.0 year. med ian = 3.1 year), and 7.2% of the sample (n = 13) had dropped out of the stu dy. Subjects remaining in ou r stu dy and subjects who withdrew did not differ in age. gender. or racial distribution; whether th ey had attempted suicide with in 2 weeks preceding their index hospitalization or had made suic ide atte m pts previously; or in the ir mean BDI, STAI Trait Anxiety. o r STAXI Trait Anger scores. Th is study is ongoing. and the yout hs are still being followed .

Instruments and Procedures ASJess11Im t of Psychiatric Disorders. Psychiatric disorder s were assessed with the Interview Schedu le for C hild ren and Adolescents (lSCA ) (Kovacs. 1985 ). The ISCA is a semistructured interview th at assesses the presence or absen ce and th e severity of psychiatric symptomatology. For each symptom assessed with the ISCA, operational criteria specify the severity levels with wh ich sympto ms are considered to be "clinically significant." Only sympto ms that are clinically significant in terms of duration, severity, and functional impairment contribute to the operational diagnosti c crite ria for psychiatric diagnoses. Interviewers using the ISCA were ment al health professional s extensively trained in the ad m inistra t ion of semistructured interview instru ments. The inrerrar er reliabil ity of ISCA symptom ratings has been found to be sat isfacto ry (Kovacs, 1985 ). The predictive validity of DSM-1I1d iagnoses assessed with the ISCA has been demonstrated (Kovacs er al., 1994. 1997). At the hospitalization assessment , th e ISCA was adm in istered to ado lescents; the inquiries of the ISCA referen ced the period of tim e pre ceding the ho sp it al ad missio n. Auxili ar y in fo rm at io n was obtained from intervi ews with parent s and other mental health profession als. behavioral observatio ns. data recorded in medi cal cha rts, and prior record s. For th e follow-up assessment s. the full ISCA was ad m inistered to both the ado lescents and ad ult info rmants (unt il subjects were age 18 or living independently). Auxiliary inform ation was obtained from schoo l. treatment, and legal records. Psych iatric d iagn oses were made o n the basis of all available infor matio n. d iscussed in co nference, and determined by consensu s. For purposes of statis t ical a nalyses, psychiatric d iagn oses were grouped into several categories. Affective disorders referred to d iagnoses o f majo r depressive d isorder, dysthym ia, bipolar d isorder, cyclothym ia, and schizoaffective d isord er. In addition, we included 3

J . A M ACAD, CHILD AD OL ESC. P SYCHIATRY. .1H:6. J UN E 1999

AD OL ESCENT SUICIDE ATT EMPTS

adolescents in this group who had a diagnosis of depressive disorder not otherwise specified: 2 of these youths had presentations resembling major depression, and the third had a presentation resemhling dysth ymia. Anxiety disorders referred to any anxiety disorder in the DSM-III-R sections "Anxiety Disorders of Childhood or Adole scence" or "Anxiety Disorders." C o nd uct/oppositional behavior disorders referred to conduct disorder and oppositional defiant disorder. Substance use disorders referred to any psychoactive subsrance abu se, psychoactive substance dependence, or psychoactive substanceind uced organic mental disorder other than nicotine dependence and nicotine withdrawal. Assessment ofSuicide Attempts. Suicide attempts were assessed with standardized ISCA questions (e.g., "H ave you ever thought about killing yourself?") and corresponding predefined rating scales. Results obtained from an interrater trial of 46 cases indicated that interrater agreem ent for the ISCA item regarding suicide attempts (computed as Cohen K) was 1.00 (M. Kovacs, unpublished manuscript, University of Pittsburgh, 1981). In accordance with definitions proposed by a National Institute of Mental Health Task Force (Resnik and Hathorne, 1973 ), a suicide attempt was defined as a self-inflicted, completed, life-threatening act that is associated with psychologi cal intent to end one's life and has the potential of resulting in ph ysical injury or harm . Self-destructive beh avior was classified as a suicide attempt if it was associated with any desire to die , regardless of multiple motives or ambivalence associated with the act. Attempts that were sto pped before the y were execut ed were considered to be suicidal ideat ion rather than suic ide attem pts. Self-harm behaviors not associated with intent to kill oneself were not considered to be suicide attempts. This definition of suic ide attempts is similar to th at used by Goldston et al. (1996, 1997, 1998) and Kovacs er al. (1993). Suicide Completions. To determine whether sub jects with whom we had lost contact had died . we searched the publicl y available National Death Index. Assessment of Severity of Depression. Anxiety. and Anger. Adolescents also were administered the BDI (Beck et al., 1988), the STAI (Spielberger et al., 1983), and th e STAXI (Spielberger, 1988). Youths typically began completing self-report questionnaires on their first or second day on the inpatient unit. Questionnaires were read to youths who had learning disabilities or difficulties with reading. The BDI is a widely used self-report inventory for the assessment of the severity of depressive symptomatology over the preceding week (Beck et al., 1988) . Among adolescent inpatients, the BDI has been found to be an internally co nsistent scale (Cronbach coefficient ex = .79) with moderate 5-day test-retest reliability (Strober et al., 1981). The STAI is a reliable and valid self-report measure of state and trait anxiety (Spielberger et al., 1983). In this study. we focused o n the construct of trait anxiety. For high school stu dents, normat ive data for th is measure have been reponed. and the measure has been found to have excellent 3D-day test-rete st reliability and a high degree of internal consistency (Spielberger et al., 1983). The STAXI is a self-report measure that assesses trait anger. state anger, and the expression of anger (Spielberger, 1988 ). In this study, we focu sed on the construct of trait anger. Among adults, stu d ies have suggested adequate test-retest reliabil ity. internal consi sten cy, and construct validity for trait anger; normative data also have been reported for high school stu dents (Spielberger, 1988). Assessment of Sociodemograpbic Variables at Hospitalization. Sociodemographic variables including gender. race, and age at hospitalization were recorded d irectly from medical records . Occupation and edu cation of parents (for determining SES) were assessed with the Follow-up Information Sheet , a data entry form modeled after the

J.

AM . ACAD. CHILD ADOL ES C. PSYCHIATRY. 58:6. JUNE 19 9 9

Intake Information She et developed by M . Kovacs (unpubl ished manuscript, University of Pittsburgh. 1982).

Statistical Methods Risk for a first po srho sp iralizarion suicide attempt and facto rs modifying that risk were assessed with Kaplan-Meier estimates of the survival distribution (Kaplan and Meier, 1958), a Weibull param etric hazard funct ion , and C ox proportional hazards models (Cox and Oakes, 1984). These methods use all available data, regardless of varying numbers and timing of sub ject int erviews. Subjects not mak ing a suicide attempt were censored at their most recent interview. All covariares in models reflected measurements taken at the baseline hospitalization. Adequacy of the We ibull distribution for describing the hazard function was examined using log-log plots of the estimated survival distribution. The hypothesis of a constant versus Weibull hazard was tested by fitting a parametri c Weibull model using the maximum likelihood method (Cox and Oakes, 1984). For the Cox models, assumptions of the proportionality and constancy of effects over time were tested using the correlation of the Schoenfeld residuals and event times (Schoenfeld, 1980; T herneau er al., 1990). The effects of covariates were examined both as un ivariate predictors and in models with multiple predictors. Except as noted under "Results," age, gender, and race were always included in model s with multiple predictors. For analyses in which diagnostic group ings. psych iatric comorbidiry, and co nt inuous psychiatric measures were considered, the number of previous suicide attempts was also included in predictive models. All model s involving interactions also controlled for corresponding main effects.

RESULTS

Suicidal Behavior and Psychiatric Diagnoses at Index Hospitalization

Of the 180 adolescents in the sample, 105 (58.3%) did not have a history of suicide attempts before hospitalization. Of the remaining adolescents, 41 (22.8%) had attempted suicide once , 21 01.7%) had attempted suicide twice, and 13 (7.2%) had attempted suicide between 3 and 7 times before their index hospitalization. Of the 75 adolescents with a history of attempts, 42 (56%) had attempted suicide within 2 weeks preceding their hospitalization. Of the 138 adolescents who did not attempt suicide within 2 weeks preceding their hospitalization , 60 (43.5%) described suicidal ideation at the time of their admission. Of these, 44 (73 .3%) reponed a specific plan for suicide. The distribution of psychiatric diagnoses at index hospitalization for the longitudinal sample was similar to that of the larger sample from which the adolescents were recruited (Goldston et al. 1998). In this group of 180 adolescents, 91 (50.5%) had affective disorders, 32 07.8%) had anxiety disorders, 97 (53.9%) had conduct

663

GOLDSTON ET AI..

or oppositional disorders, and 31 (17.2%) had substance use disorders at hosp italization. The rates of preh ospitalizarion suicide attempts were 50 .5 % , 46. 9% , 32.0%, and 45 .2%, respectively. among adolescents with affective, anxiety, conduct/oppositional, and substance use disorders. The relationship between index diagnoses and prehospiralizarion suicidal behavior ha s been reported elsewhere (Goldston et al., 1998) . Posthospitalization Suicide Attempts

By the cutoff date for these analyses, 37 adolescents had made at least one suicide attempt after hospital discharge, but no adolescents to our knowledge had killed themselves. Eighteen (48.7% ) of the youths had attempted suicide by overdose, 14 (37.8 %) by stabbing or cutting, 3 (8.1% ) by hanging or other means of strangulation, 1 (2.7%) by running in front of a car (an incident in which he was hit), and 1 (2.7%) by inserting multiple straight pins and pieces of graphite from a pencil under his skin in the belief that he eventually would die of lead poisoning or infecti on. Periods of Risk After Hospitalization

The su rvival distribution curve (with 9 5% confidence limits) for the risk over time for a first suicide attempt

after hospitalization is presented in Figure 1. An exponential (constant) hazard for the risk over time after hospitalization was rejected in favor of the Weibull distribution hazard function shown in Figure 2 with 95 % co nfidence limits (z = 2.95 , P = .003) . By visual inspection, the first 6 months to 1 year after hospitalization represents a particularl y high-risk time for att em p ted sui cide. Age at hospitalization, gender, race, and SES were not relat ed to the time until posthosp iralization suicide attempts. In contrast to the sociodemogr aphi c va riables. a greater number of suicid e att em pts prior to ho spital ization wa s related to earlier posthospitalizarion suicide attempts (b = 0.332, exp[b] = 1.380, serb] = 0.121, P = .008). The effect of increasing previous sui cide attem p ts is seen in the Kaplan-Meier estimate of sur vival until first su ic id e at te m p t stratified by 0 , 1, or > 1 prior attempts in Figure 3. As can be seen , the survival func tions for youths with a single suicide attempt and with no prior attempts were similar. However, repeat suicide attem pters were at tw ice th e risk (i.e., 102% greater risk) for posthospiral izat ion suic id al beh avior than the other yout hs com bined (p = .0 5 1). The cffects of prior number of suicid e atte m p ts seen in thi s figure d id not differ signifi cantly over the course of the ob servation period.

C!

C>

0

•. ...

i

__

_

_

S

a. E

.,

= .,

....•...........

..._

...

<0

«1: 0 "C

'u '5

(/)

,..

:; ci 0

€

~

<: 0

€

CD

ae 0

11.

ll)

ci

v ci

0

2

3

4

5

Years Since Discharge Fig. 1 Proportio n of formerly hospitalized ado lescents (with 95 % hound s) remaining free of suicide all emprs d uring rhe firsr 5 years afrer discharge.

664

J. AM . ACA D. CH IL D ADOL ESC . PSYCHI ATRY . .\H : (" JU N E 19 99

ADOLESCEN T SU IC ID E AT T EMPTS

0 C'l

ci

10

C\l

ci

0 C\l

ci c: 0

·U

c:

:::J

U.

10

"E ci
~

J: 0

ci

10

0

ci

0

c:i 0

2

3

4

5

YearsSinceDischarge

Flg.2 Parame tric hazard fun crio n depicting risk for ma king a first posrhos piralizarion suicide att empt d uring th e first 5 years ali er discharge.

~

,,

en

c:i

L. _ _ ,

, L ,

,

III

a. E

~

L ...... ,

,, ---,

co ci

~

'-- - _ ._-- ,

CD

,

"0

:§ :::J (/)

._.....+...... ..... +-•...

,

... +++-- -~--

- - - ...

r-,

'5 ci 0

s:

"" ;: c: .~

8.

,,, --- ,, ,,- -+ - - - - ...... -

- - I

~

- -- - - + -

~

-+- - - - - - - -+-+-+ -

- - -+-


ci

e

a,

10

ci

-e ci 0

2

3

4

5

YearsSince Discharge

Fig. 3 Proport ion of formerl y hospi ta lized adoles cents (stratified by nu mb er of pre hos pi ralizario n su icide attempts) remai ni ng free of suic ide attem pts during the first 5 years afte r d ischarge.

J. AM . ACAD . C H I l. D AD O I. ESC. PSY CH IAT RY. .18 : 6 , J UN E 19 99

665

GOLDSTON ET A I..

Posthospitalization Attempts as a Function of Recency of Past Attempts and Suicidal Ideation For the analysis regarding recency of attempts, we focused only on adol escents who had a history of (any) suicide attempts prior to their hospitalization (n = 7 5). Within thi s subset of adolesce nts, att empting su icide within the 2 weeks immediately preceding hospitalization was not predictive of tim e until first posrho spitalization attempts. For th e anal yses regarding suicidal ideation, we examined both adol escents with a previou s history of suicide attempts , bur no recent attempts, and adole scents with no history of suicide attempts (n = 138) . For this combined sam ple, suicidal ideation at hospitalization was not predictive of time until first posrhosp iralizarion attempts. The interaction term between suicid al ideation and history of suicide attempts also was not significant. Psychiatric Diagnoses and Posthospitalization Suicide Attempts

Continuous Psych iatric Measures and Suicide Attempts In the univariate C ox models, severity of depressive sym pto mato logy was strong ly related to risk for suicide attem pts (b = 0.0 39 , exp[b] = 1.039, serb] = 0.012, P = .00 I ). Severity of trait anxiety also was related to risk for posrhospiralizarion att empts (b = 0.033, exp[b] = 1.033, serb] = 0.013, P = .010). Trait ang er was not related to

suicidaliry, In the Cox models with sociodemographic variables, diagnoses, and past suicide attempts included as covariares, severity of depressive sym pto matology was still found to be strongly related to risk for suicide att empts (b = 0.048 , exp[b] = 1.049 , serb] = 0.015 ,p = .002). Likewise, with all of the covariares included, trait anxiety also was related to risk for suicide att em pts (b = 0.032, exp[b] = 1.032, serb] = 0.016, P = .048). Trait anger was not related to posthospitalization attempts. Differential Predictive Utility of Psych iatric Variables as a Function of History of Suicide Attempts

The diagnostic groups by themselves were nor significantly related to later suicidal behavior. However, after controlling for sociodemograph ic variables and prior suicide attempts, we found that conduct and oppositional di sorders were related to po sthospitalization sui cide att empts (b = 0.728, exp[b] = 2.071 , se rb] = 0 .36 7, P = .047). Supplementary anal yses examining interaction s of diagnostic groups with gender did not indicate th at diagno sis effects were moderated by gender. Moreover, the effect of prior suicide attempts remained predictive, even after we controlled for th e presen ce of each of the psychiatric diagnosis groups (i.e., forcing them into the model ) a nd sociodemograph ic vari ables ( b = 0. 31 8 , exp[b] = 1.37 4 , serb] = 0.122, P = .00 9).

As shown in Figure 4, affective disorders at hospitalizatio n were more strongly predictive of suicide arrempts among indiv iduals with historie s of prior att em pts than they were for ind ivid uals who had never made a suicide arrem pt (b = 1.254, exp[b ] = 3.504, serb] = 0.562, P = .026) . Anxiety, conduct/oppositional, and substance use disorder s did not interact with history of suicide attempts. The interaction term between trait an xiety and past att em pts was not significant . However, th e interaction between severity of self-repo rted depression and past attempts was significant (b = 0.026, exp[b] = 1.026, serb] = 0.010 , P = .007). As seen in Figure 5, the relationship between depressive sympto ms and later suicidality was more pronounced amo ng adole scents with a history of suicide attempts.

Comorbid Psych iatric Conditions

DISCUSSION

Two patterns of psychiatric comorbidity were specifically examined: affective disorder with conduct/oppositional disorder and affective disorder with substance use disord er. To exam ine wh ether these co mo rb id condition s predicted later suicidal behavior, we exam ined (in separate C ox proportional hazard s models) th e int eraction terms of affective di sorder X conduct/oppositional behavior disorder and of affective disorder X substance use d isorder. Ne ither pattern of psychiatri c comorbidity was pred ictive of posrh ospitalization attem pts.

In this pro specti ve, naturalistic, repeated-assessments stu dy of 180 adolescents after psychiatric hospitalization, we examined both the risk over tim e and the psychiatric pred ictors of posrhospiralization suicide attem pts. We estimated from the Kaplan -Meier curve that approximately 7% of adolescents make suicide atte mpts within the first 6 months after their discharge, 12% attempt suicide within the first year after di scharge, and 25% attempt suicide within the first 5 years after hospitalization. In comparabl e sam ples, Brent er al. (l993 a) and

666

,

AM . ACAD . C1I I U l ADO LESC . PS YC Ii IAT RY. .IH :(, . J U N E 1') ')<)

ADOLESCENT SUICIDE ATTEMPTS

Co!

~'--i-'

--+++,

\ L L.,-+-,

CJ)

~

~ - ~-++++-

~

0

L_ III

0. E Q)

= <

1 L

CI)

1 L_

0

...........+

J':"i-+': ;::;+·+··..:~·.:·:.:~:..:·+·.:·*·:.:·+·:.:·:.:·.:·::·:.;:±t.+·.:·~·.t:.::'!'+-.:.:t·

I L---

:2:::>

L_

r-,

-+ -+--

1 4+----+-+++-

1 1

s:

'-_I

~

8.

-t++-+I-++ -

1

'5 0 0

1:

.............. +H

L_]

Q)

c: .2

,, ,, - +- - -

1

"0

(/)

+--+---i--+-\

........++-

1

'-_I I

to

0

e

a.

Previous Attempts and Affective Disorder ll)

No Previous Attempt, No Affective Disorder No Previous Attempt, Affective Disorder Previous Attempt, No Affective Disorder Previous Attempt, Affective Disorder

0

o

2

3

4

5

Years Since Discharge

Fig. 4 Risk tor suicide attempts during the firsr 5 years after discharge as a function of aftective disorders and ptevious attclnpts.

1- -, CJ)

0

~

0.

L_

CI)

~ 0

,-+-+-

-

-+-

, I

---H-,

1 L

,

1 L+-

Q)

"0

1

L_]

:~

:J (/)

1..--++--++-- 1

\

III

E Q)

,,

1'1

L-+--

r-,

'5 0 0

1 1

--I

f;

:i:

4+-----+--+H-l

c: 0 to t 0

1

1

--I

8.

e a.

1 1

Previous Attempts and BDI Scores

--I 1

ll)

No Previous Attempt, BDI <16 No Previous Attempt, B0116+ Previous Attempt, BDI <16 Previous Attempt, B0116+

0

o

2

1 -----~~------~--+----+--+~

3

4

5

Years Since Discharge

Fig. 5 Risk for suicide attempts during the first 5 years after discharge as a function of Beck Depression Inventory (BDl) scores and previous attempts.

J. AM. ACA D. C H I1.D AI)()LFSC. PSYCH IATRY. -'H:(,. JU N E 1~~~

667

GOLDSTON ET AI..

King er al. (1995) found that 10% and 18% of previously hospitalized teenagers, respectively, attempted suicide within 6 months of discharge. Taken together, these data suggest that the first year after psychiatric hospitalization is a very high-risk period for attempted suicide among adolescents. Results from the current study, however, extend previous findings by showing that the risk for suicide attempts diminishes somewhat (but does not disappear) during and after the first year after hospital discharge. Adolescents with a history of multiple attempts prior to hospitalization seemed to be at especially high risk for additional suicidal behavior. For these youths, the risk for posthospitalization suicide attempts was twice as high as that of youths with no suicide attempts or with only a single prior attempt. Previous suicidal behavior has long been thought to be one of the strongest predictors of later suicide attempts in both adolescents and adults (e.g., Brent et aI., 1993a; Leon et al., 1990; Pfeffer et al., 1993). Indeed, among adult inpatients, each suicide attempt has been found to be associated with a 32 % increase in the relative risk for later suicide attempts (Leon et al., 1990). Given the relationship between history of suicidal behavior and future attempts, it is notable that a significant number of adolescents without a history of suicide attempts prior to their psychiatric hospitalizations made their first suicide attempts after their discharge. Specifically, 3.8% of previou sly nonsuicidal adolescents made attempts within 6 months, 7.8% attempted suicide within 1 year, and 25.6% attempted suicide within 5 years after hospitalization. The 6-month and 1-year rates among previously nonsuicidal youths are similar to the rate of attempted suicide for previously nonsuicidal youths reported by Brent et al. (1993a) , and they indicate that the high risk for suicidal behavior after hospitalization is not limited to adolescents with previous suicidal behavior. The majority of youths in psychiatric hospital settings have multiple stresses, poor coping skills, and significant functional impairment, and the fact that they have not attempted suicide in the past should not be construed as indicating a lack of risk for posthospiralization suicidal behavior. Neither the timing of adolescents' suicide attempts before hospitalization nor the presence of suicidal ideation among youths without recent attempts was related to subsequent suicidal behavior. Although the recency of prior su icid e attempts and the presence of suicidal thoughts have been co nsid ered to be indicators of

potential self-harm in clinical settings, data from this sample suggest that such factors may not be predictive of long-term outcome, at least in a multiply disturbed inpatient sample. Alternatively, it is possible that it is not proximity of past suicide attempts or suicidal thoughts per se, but rather the combination of these factors and other risk factors that portend the greatest posrhospiralization risk. Studies focusing on the predictive utility of risk factors among different groups of inpatient youths are su rprisingly rare and are greatly needed to further clarify these issues. In terms of psychiatric risk factors, we found that affective disorders were predictive of posthospitalization suicide attempts, but only when accompanied by a prior history of suicide attempts. Affective disorders at the time of hospitalization did not have prognostic utility in the absence of other cont ribut ing risk factor s. Prospective studies of clinic ally ascertained and mixed samples of ch ild ren and adolescents have found that major depre ssion (Brent er al., 1993a; Myers et aI., 1991a), dysthymic disorder (King et al., 1995), and affective disorders in general (Brent et al., 1993a; Kovacs et aI., 1993; Pfeffer et al., 1993) are predictive of later suicidaliry, In most studies, however, the potentially interacting effects of past suicide attempt status and baseline affective disorder have not been examined. In one study addressing this issue , Pfeffer et al. (1993) found that suicid aliry status did not interact with other risk factors in predicting long-term risk for suicide attempts among prepubescent children. However, suicidal children were more likely to have subsequent mood disorders, and youths with subsequent mood disorders were more likely to att empt suicide. Brent et al. (1993a) also found that hospitalized adolescents "with a history of both suicidality and depression were at greatest risk for clinically significant suicidality during ... follow-up" (p. 101). Conduct and oppositional defiant disorders by themselves were also not predictive of suicidal behavior. However, the presence of conduct/oppositional disorder was related to increased risk in analyses controlling for past suicidal behavior and sociodemographic variables. Findings regarding the predictive utility of conduct disorder have not been consistent across studies, in part because of different sampling strategies and methodological differences. Particularly in a high-risk inpatient sample, the presence of any single psychiatric disorder at a single point in time (hospital admission) may not convey the same prognostic information as the same diag-

668

J . AM. ACAD. CHILD A [) O L E SC. P SYCHIATRY, .18:(' , JUNE 1999

ADOLESCENT SUICIDE ATTEMPTS

nostic information in a non-clinically referred sample. The combination of baseline psychiatric disorders with other risk factors, and the course and stability over time of psychiatric disorder, likely provide more reliable and accurate prognostic information about later suicidal behavior than index psychiatric diagnoses in isolation. Results from this sample indicated that comorbidity with affective disorder was not predictive of later suicide attempts after we controlled for the independent (main) effects of affective and nonaffective disorders. In an outpatient sample, the presence of substance and/or conduct disorders was noted to increase the risk for subsequent suicidality beyond that associated with affective disorders alone (Kovacs et al., 1993). Likewise, comorbid affective and substance use disorders more commonly characterize adolescent suicide victims than matched community controls (Brent et al., 1994; Shaffer et al., 1996) and matched community controls with a history of affective disorder (Brent et al., 1994). In contrast, there is little evidence to suggest that comorbid affective and nonaffective disorders significantly increase the risk for later suicidal behavior in an inpatient sample often typified by multiple diagnoses. Even after we controlled for presence of psychiatric disorders and history of past attempts, severity of depressive symptoms was predictive of later suicidal behavior. Prospective studies of juvenile suicidal behavior have typically focused on nosological categories such as affective disorders, rather than severity of depressive symptoms (e.g., Brent et al., 1993a; Kovacs et al., 1993; Pfeffer et al., 1993). One notable exception is a study by Myers et al. (1991a) in which severity of depressive symptoms was found to be predictive of "suicidaliry status" (rated on a continuous scale) in both a diagnostically heterogeneous group of adolescents and in a subsample of adolescents with major depression. The severity of depressive symptoms during adolescence is predictive of functioning in a variety of areas in young adulthood (Devine et al., 1994) and appears to convey significant prognostic information over and beyond that simply associated with affective disorder diagnoses. Trait anxiety also was predictive of posthospitalization suicide attempts, independent of psychiatric diagnoses. These findings complement cross-sectional studies that have characterized adolescent suicide attempters in hospital settings (e.g., Ohring et al., 1996), and repeat suicide attempters in particular (Goldston et al., 1996), as evidencing significant trait anxiety. Anxiety-related symp-

toms also have been found to be predictive of suicide in adults and are thought to be related to one of the clinicalbiochemical pathways associated with suicide (Fawcett et al., 1997). The final question of this study focused on the differential predictive utility of psychiatric variables among individuals with and without prior histories of attempts. In this formerly inpatient sample, we indeed found that both affective disorders and severity of depressive symptoms were more strongly predictive of posthospitalization attempts among individuals with histories of suicide attempts than among individuals without such histories. There are a variety of possible explanations for this effect. However, we would suggest that once individuals have actually made suicide attempts and not just thought about it, they traverse a certain behavioral "threshold." That is, such individuals know they are capable of making a suicide attempt and suicide attempts are unequivocally in their behavioral repertoire. Such individuals will be at increased risk for engaging in similar behavior in the presence of internal and environmental cues similar to those present at the time of earlier suicide attempts. With regard to the current study, affective disorders in particular often have a prolonged and recurrent course and are likely to provide reoccurring depressive states similar to that associated with earlier suicidal behavior. Recurrent suicidal behavior may be especially likely to recur if individuals received relief from an undesirable inner state (i.e., amelioration of pain or catharsis) or if their earlier suicide attempt prompted a desirable change in the environment. Future longitudinal studies are needed to examine whether other purported risk factors for suicidal behavior also have more predictive utility among individuals with histories of suicide attempts than among individuals without such histories.

J. AM. ACAD. CHILD ADOI.ESC. PSYCHIATRY, .18:6, JUNE 1999

669

Limitations This study defined recent suicide attempts as attempts occurring within 2 weeks of hospitalization and did not focus on self-destructive behavior without intent to die. In addition, although mirroring the community from which they were drawn, this clinically referred sample was 80% white. Studies using different definitions of suicidal behavior or focusing primarily on minority populations may yield different results. In addition, subjects in this study were recruited from a single inpatient setting, and the findings may therefore not be generalizable to community or other clinical settings.

GOLD STON ET AI..

Conclusions and Clinical Implications

to further clarify the interaction, stability, and covariation of these risk factors over time.

The findings from this study, if replicated, are informative with regard to prevention and intervention efforts with high-risk hospitalized adolescents. First, these data suggest that the first 6 months to 1 year after hospitalization is a particularly high-risk period for suicide attempts. Aftercare services and close monitoring are especially important during this time, as are step s to avoid the premature or unplanned discontinuation of treatment. Second, because approximately 1 in 4 youths make suicide attempts in the 5 years after they are hospitalized, all youths in such settings should be considered to be at high risk. Clinicians should continue to closely monitor the potential for self-harm among previously hospitalized youths, even in the absence of prior suicidal behavior. Third, because of the increased risk associated with a history of multiple attempts, repeat suicide attempters should be monitored carefully. Such adolescents are especially in need of ongoing intervention to promote adaptive coping strategies, to modify the propensity for suicidal behavior, and/or to effect changes in both the contextual and proximal factors that have precipitated suicidal behavior in the past (Berman and Jobes, 1995) . Fourth, affective disorders and severity of depressive symptoms are strong risk factors for suicidal beh avior among ad olescents who have already made suicide attempts. However, our findings also suggest that mood disorders and depressive symptoms should not necessarily be considered by clinicians to be indicants of later risk among hospitalized adolescents with no prior histories of suicidal behavior. Our findings likewise suggest that recency of past suicidal behavior and the presence of suicidal ideation among hospitalized youths may be of limited utility-by themselves-in predicting suicide attempts. Fifth, among hospitalized youths, trait anxiety and associated feelings of agitation increase the risk for suicidaliry over and beyond that associated with psychi atric diagno ses. Psychosocial and pharmacological interventions should include a focus on reducing such anxiety-related symptoms and fostering the sense of control that is antithetical to anxiety. In summary, it is clear that the likelihood of suicidal behavior among formerly hospitalized adolescents is multiply determined and influenced by interacting risk factors. The repeated-measures methodology of thi s ongoing longitudinal stu dy will allow us an opportunity

Andrews J . l.cwinsohn P ( 1992). Suicidal attc mp ts amo ng olde r adolescent s: prevalen ce and co -occu rrence with psych iatr ic disorders .} Am Acad Child Adolrsc Psychiatrv j I:6 55-662 Mclnrirc M (l98 .~ ). Adolescent self-poiso ni ng: a nineAngle C. O 'Brien year follow-up. D" , Brha» Pedintr 4:83-87 Bart er J. Swaback D . Todd 0 (1968). Adolescent suicide anempts: a followup study o f hosp italized pat ient s. Arch G(1I Prychiatry 19:5B-527 Beaut rais A. Joyce r. Mulder R ( 1998). Psych iatr ic illness in a New Zealand sam ple of young peopl e making seriou s suicid e atrem p ts. N Z Med ] 111:44- 48 Beck A, Steer R. Garbin M (1988), Psychometric properties of the Beck Depression Inventory: twen ty-five years of evaluation . Clin Psychol Rev 8:77- 100 Berm an A, Johes 0 (199 5), Suicide prevention in adolescents (age 12-18) . Suicid« Lift Threat Brba» 25: 143-154 Brenr D. Kolko D. Wartdla M et al. (1993a ). Adolescent psych iat ric inpa ricn rs' risk for suicide arrem pr at 6- mont h follow-u p. } Am Acad Child Adolesc Psvcbiatry .U :9 5- 105 Brent D. Perper J. Mo ritz G et al. (l 9 ')jb ). Psychiatric risk factor s for adolescent suicide: a case-co ntro l study. } Am Acad Child Adolrsc P~ychiatry .U :52 1- 529 Brent D . Perper J. Mnritz G . Baugher M. Sch weers J. Roth C ( 1994) . Su icide in affeClively ill adolescents: a case-control study.} Ajftct Disords I: 193-202 C ox D . Oa kes D (1984). Analysis ofSuruiualData. New Ynrk: C ha pman and H all Devin e D. Kernprom T. Forehand R (1994). Adolescent depressed mood and you ng adult functioning: a longirudinal stud y } Abnorm Child Psycho! 22:629- 64 0 Fawcett J. Busch K. Jacobs D . Kravitz H . Fogg L (1997). Suicide: a fourpathway clin ical-bioch em ical model . Ann N Y Acad Sci 856:288-301 G old sto n D . Daniel S. Reb ou ssin B. Reb ou ssin D. Kelley A. Fraz ier P (1998). Psychiatri c d iagnoses amo ng previou s suicide attemp ters. firsttim e att em p ters. an d rep eat atrem p ters on an ad o lesce nt inp ati ent psych iatr y un it.} Am Acad Child Ado/esc Psychiatry 37 :924-932 Gold ston D . Daniel S. Reboussin D . Kelley A. levers C. Brun sretter R (1996) , First-t ime suicide an em pters, repeat ane m plers, and previou s an ernpters on an adolescent inpatie nt psych iatr y un it.} Am Acad Child Ado/esc Psychiatry .~ 5 : 6 3 1-6j9 Goldston' D. Kelley A. Reboussin D er al. ( 1997). Suicidal ideati on and beh avior and noncompliance with the medi cal regimen am ong d iabetic ado lcsccnts. j' Am Acad Child Adolesc Psychiatry 36 :1528-1536 Goul d M . King R, Greenw ald S et al. ( 1998). Psychopatholo gy associated with suicida l ideat ion and an em p" among ch ildren and ad olescent s.} Am Acad Child Adolesc Psychiatry 57 :9 15-9B Harr ington R. Bredenk arnp D . Groothues C. Rutter M . Fudge H . Pickles A (1994 ). Adult ou tco mes of child hood and ado lescent depre ssion . III : links with suicidal behaviours.} Child Psycho! Psychiatry 3 5: B09-1J 19 H awron K. Fagg J. Platt S. H awkins M (1993) , Fa~tors as~ociated with su icide after parasuicide in you ng people . BM} 30(,: 1641-1 64 4 Holl ingshead A (19 57 ). 71"0 Factor Index of Social Position. New H aven , CT: Yale Un iversity Dep artment of Sociology Kap lan E. Meier P (195 8). Nonparametri c est imatio n from incomplete o bserva rio ns. YAm Stat Assoc 8:699 Kerfoot M. McHugh B (1992 ). The outcome of ch ildhood suicidal beh avior. Acta Piledopsychiatr 55 :141-145 King C. Segal H, Kaminsk i K. Naylor M . Gh aziuddin N . Radp our L ( 199 5). A prospective study of ado lescent suicida l beha vior following hospital ization . Suicide Lift Threat Brhau 25:327-338 Kovacs M (1985 ). The Int erview for Ch ildren (ISC). Psychopharmarol Bill! 2 1:991- 994 Kovacs M. Garsonis C. Pollock M. Parron e P (1994). A co nt rolled prospective study of DSM-II/ ad justme nt disorder in child hood : sho rt-term prognosis and long-term pred ict ive validity. Arch Gen Psychil1t~y 5 1:535- 54 1

670

J. AM . ACAD. CHILD AIl OLESC. PSY C HI ATRY . .lH:6. J UN E 1999

REFERENCES

i.

ADO LESCENT SU IC IDE AT T EMP T S

Kovacs M. Goldsron D. Gatson is C (1993). Suicidal behavior and childhoodon set depr essive disorders: a longitudinal investigation .] Am Acad Child Adolrsc Psychiatry 32:8-20 Kovacs M. Obrosky S. Gatsonis C. Richards C (1997), First-episode major depre ssive and dysthymi c disorder in childho od : clinical and sociodernographic factors in recovery.] Am Acad ChildAdolesc Psychiatry36:777-784 Leon A. Fried man R, Sweeney I. Brown R, Mann j (1990), Statistical issues in the ident ification of risk facrors for suicidal behavior: the application of survival analysis. Psychiatry Res 31:99- 108 Mcintire M. Angle C. W ikotT R. Schlicht M (1977). Recurrent adolescent suicidal behavior. Pediatrics 60:60 5- 608 Myers K. McCauley E, C alderon R, Mitchell I. Burke P. Schloredt K (1991a), Risks for suicidaliry in major depressive disorder.] Am Acad Child Adolesc Psychiatry 30:86-94 Myers K. McCauley E. Ca ldero n R. Treder R (1991h). T he 3-year longirudinal course of suicidaliry and predictive factors for subsequent suicidaliry in youth s with major depressive disorder.] Am Acad Child Adolesc Psychiatry 30 :804-810 Ohring R. Apter A. Ratzoni G . Weizman R. Tyano S. Plutchik R (19% ). State and trait anxiety in adolescent suicide arrernprers. j' Am Acad Child Adolesc Psychiatry 35:154-1 57 pfetTer C. Klerman G. Hurt S. Kakum a T, Peskin l - Siefker C (1993), Suicidal child ren grow up: rates and psychosocial risk facrors for suicide attempts durin g follow-up.] Am Acad Child Adolesc Prychiatry 32: 106-11 3

Resnik H. Ha thorne B (1973) . Suicide Preumtion ill the 70s. Washingron , DC: US Government Printing Office Rorheram-Borus M (1989). Evaluating suicidal youth in communi ty settings. In: Innovations in Clinical Practice: A Source Book. Vol 8. Keller P. Heyman S. eds, Sarasota, FL: Professional Resource Exchange Schoe nfeld 0 (1980) . C hi-squared goodne ss of fit tests for proport ional hazards regression models. Biometrika 67 :145-153 ShatTer 0, Gould M, Fisher P et al. (19% ), Psychiatric diagnoses in child and adolescent suicide. Arch Gm Psychiatry 53:339- 348 ShatTerD. Vieland V. Garland A. Rojas M. Underwood M. Busner C (1990). Adolescent suicide a!tem pters: response to suicide preventi on programs. ]AMA 264 :3 15 1-3 155 Spielberger C (1988). State-Trait Angrr Expression lnoentory Research Edition: Profnsional Manual. O dessa, FL: Psychological Assessment Resources Spielberger C. Gor such R. Lushene R. Vagg P. jacobs G (1983). MarlllalfOr the State-Trait Anxiety Inventory. Palo Alto, CA: Consulting Psychologists Press Spiriro A, Plummer B. G isperr M er al. (1992 ). Adolescent suicide attempts: outco mes at follow-up. Am] Orthopsychiatry 62:464- 468 Strober M, G reen j , Carlson G (198 1), Utiliry of the Beck Depression Invenrory with psychiatrically hospitalized adolescents. ] Consult Clin Psycho! 49 :482-483 T herneau T, Grambsch P. Fleming T (1990). Martingale-based residuals for survival models. Biometrika 77:147- 160

Sleep and Daytime Behavior in Children With Obstructive Sleep Apnea and Behavioral Sleep Disorders. judith Owens, MD . MPH. Lisa Opipari, PhD , Chanrelle Nob ile, BA, Antho ny Spi rito, PhD

Objective: The purpo se of th is stu dy was: I) to exam ine both bedtime sleep behaviors and dayt ime beha viors associated with daytime sleepiness in a gro up of children with a pr ima ry med ical sleep disord er (obstructive sleep apnea synd rome IOS ASJ) compared with a group of children with a primary behavioral sleep disorder (BSD) (limit serring sleep disorder or sleep ons et association disorder); and 2) ro investigate the impact of a comorbid BSD on sleep and daytim e behavioral conseq uences of O SAS. Methods: Childr en referred to a ped iatric sleep disord ers clinic during a 3-year period wirh a primary diagn osis of either polysomnographicallyco nfirmed OSAS (n = 100) or a BSD (n = 52) were compared on several parent report measures assessing the following domains: sym pto ms o f sleep d isord ered breathing, other sleep beh aviors (p rima rily parasom n ias), bed time behaviors, an d externalizing daytime behavior p robl em s. The O SAS sample was then d ivided into a pure OSAS gro up (11 = 78) and an OSA S plu s a behavio ral sleep d iagno sis group (n = 22) based on th e presence or absen ce of dela yed sleep onset andlor prol on ged nighrwakings and com pared o n th e parent-repore system domains. Results: Almo st one-quarter of the O SAS gro up had clin ically sign ificant beh avioral sleep problems, primarily bedtime resistance . in addi tion to OS AS. Bedt ime resistan ce was associated wit h a sign ificantl y sho rtened sleep duration in both th e BSD and OSAS-BSD groups . Alth ou gh th e OSAS -BSD group had less Severe disease, as defined by polysomnographic variables, than the pur e OSAS group, the y were rated by their parent s as havin g more da ytime externalizing beh avior problems associated with da ytim e sleepi ness. Conclusions: T he results of th is stu dy suggest th at evaluation for co mo rbid BSD sho uld be done in all child ren present ing with symptoms of OSAS. The coexistence of such BSDs may contribute signi ficant ly to sleep deprivation . and thus to beh avioral manifestation s of daytime sleepin ess in these children. Pediatrics 199R;102: 117R-llR4 ; reproduced by permi ssion of Pediatrics

J. AM . AC A D . C H I L D AD O LE Sc:. PSYCHIAT RY. 5 8 :6 . JUN E 1999

671